plugin

type ClusterCache struct {
	Jobs         map[api.JobID]SchedulerJob
	Nodes        map[string]NPUNode
	Tors         *TorList
	SuperPodInfo *SuperPodInfo
}

type CommonNode struct {
	Name           string
	Capability     map[v1.ResourceName]float64
	Allocate       map[v1.ResourceName]float64
	Idle           map[v1.ResourceName]float64
	Tasks          map[api.TaskID]*api.TaskInfo
	BaseDeviceInfo string
	// node annotation and device info + switch info + node info
	Annotation map[string]string
	Label      map[string]string
	Address    string
	SuperPodID int32
	// contains filtered or unexported fields
}

type ConfigParameters struct {
	StaticParameters
	DynamicParameters
}

type DevFaultInfo struct {
	LogicId       int32
	Status        string
	Policy        string
	InitialPolicy string
	ErrorCode     []int64
	ErrorCodeHex  string
}

type DynamicParameters struct {
	PresetVirtualDevice bool
	GraceDeleteTime     int64
	SuperPodSize        int
	ReservePodSize      int
}

type FaultHandler interface {
	Execute(*ScheduleEnv, *framework.Session) error
	CheckNodeNPUByTask(*api.TaskInfo, *NPUNode) error
	ScoreBestNPUNodes(*api.TaskInfo, map[string]float64)
	UseAnnotation(*api.TaskInfo)
	PreStopAction(*ScheduleEnv) error
}

type JobScheduleInfoRecorder struct {
	// PodScheduleFlag key is job uid, value is job schedule type, true is pod scheduling, false is job scheduling
	PodScheduleFlag map[api.JobID]bool
	// ResetCMSetFlag flag to record rescheduling job has been set reset cm
	ResetCMSetFlag map[api.JobID]struct{}
	// PendingMessage record different job pending reason and unschedulable nodes
	PendingMessage map[api.JobID]PendingReason
	// ServerListRecordFlag flag to record tor affinity job has been record server list to logs
	ServerListRecordFlag map[api.JobID]struct{}
}

type NPUNode struct {
	CommonNode
	VNode
}

type NodeJobInfo struct {
	NodeIp   string
	NodeName string
	JobName  []string
}

type OwnerInfo struct {
	v1.OwnerReference
	Annotations map[string]string
	Replicas    *int32
}

type PendingReason map[string]sets.String

type PolicyBuilder = func() SchedulerPluginNeed

type ScheduleCache struct {
	// special, name, value
	Names, Namespaces map[string]string
	Data              map[string]map[string]string
}

type ScheduleEnv struct {
	JobScheduleInfoRecorder
	ClusterCache
	FrameAttr   VolcanoFrame
	OutputCache ScheduleCache
}

type ScheduleHandler struct {
	NPUPlugins    sets.String
	PolicyBuilder PolicyBuilder
	FaultHandle   FaultHandler
	ScheduleEnv
}

type SchedulerBaseAttr struct {
	// contains filtered or unexported fields
}

type SchedulerJob struct {
	util.SchedulerJobAttr
	UnscheduledReason

	JobReadyTag *bool
	SuperPods   map[string][]SuperNode
	Owner       OwnerInfo
	// contains filtered or unexported fields
}

type SchedulerPlugin interface {
	SchedulerPluginBase
	SchedulerPluginNeed
}

type SchedulerPluginBase interface {
	GetPluginName() string
	SetPluginName(string)
	GetAnnoPreVal() string
	SetAnnoPreVal(string)
	GetAnnoName() string
	SetAnnoName(string)
}

type SchedulerPluginNeed interface {
	// ValidNPUJob Valid the job part of npu scheduler policy, if not, disallowed.
	ValidNPUJob() *api.ValidateResult
	CheckNodeNPUByTask(*api.TaskInfo, NPUNode) error
	ScoreBestNPUNodes(*api.TaskInfo, []*api.NodeInfo, map[string]float64) error
	UseAnnotation(*api.TaskInfo, NPUNode) *NPUNode
	ReleaseAnnotation(*api.TaskInfo, NPUNode) *NPUNode
	PreStartAction(ssn *framework.Session) error
	InitMyJobPlugin(util.SchedulerJobAttr, ScheduleEnv) error
}

type Server struct {
	IsUsedByMulJob bool   `json:"-"`
	NodeRank       string `json:"-"`
	IP             string `json:"server_ip"`
	Count          int    `json:"npu_count"`
	SliceId        int    `json:"slice_id"`
	Jobs           sets.String
	CurrentJob     *api.JobID
	Name           string
}

type StaticParameters struct {
	OnceInit              *sync.Once
	UseClusterD           bool
	SelfMaintainAvailCard bool
	NslbVersion           string
	SharedTorNum          int
	IsFirstSession        *bool // scheduler first session message is unreliable
	ForceEnqueue          bool
}

type SuperNode struct {
	Name       string
	SuperPodID int32
}

type SuperPodInfo struct {
	SuperPodReschdInfo        map[api.JobID]map[string][]SuperNode // cache super pod re-schd info
	SuperPodFaultTaskNodes    map[api.JobID][]string               // cache fault task nodes info
	SuperPodMapFaultTaskNodes map[api.JobID]map[string]string      // cache task and nodes for stage2
}

type TaskDevInfo struct {
	RankId int
	DevFaultInfo
}

type TaskResetInfo struct {
	RankList            []*TaskDevInfo
	UpdateTime          int64
	RetryTime           int
	FaultFlushing       bool
	GracefulExit        int
	RestartFaultProcess bool
}

type Tor struct {
	FreeServerCount int
	IsHealthy       int
	IsSharedTor     int
	Id              int       `json:"tor_id"`
	IP              string    `json:"tor_ip"`
	Servers         []*Server `json:"server"`
	Jobs            map[api.JobID]SchedulerJob
}

type TorList struct {
	TorLevel string `json:"-"`
	Version  string `json:"version"`
	TorCount int    `json:"tor_count"`
	Tors     []*Tor `json:"server_list"`
	// contains filtered or unexported fields
}

type TorShare struct {
	IsHealthy   int
	IsSharedTor int
	NodeJobs    []NodeJobInfo `json:"nodes"`
}

type UnscheduledReason struct {
	Reason PendingReason
	*sync.Mutex
}

type VChip struct {
	PodMap map[string]*v1.Pod
	ID     []string
	// Name Ascend910-0
	Name string
	// Kind Ascend910/Ascend310/Ascend310P
	Kind        string
	IsDual      bool
	Unstable    bool
	CoreNum     int
	SegmentFlag bool
	TotalRes    util.VResource
	UsedRes     util.VResource
	FreeRes     util.VResource
}

type VNode struct {
	// Chips map chipID to VChip class
	Chips map[int]*VChip
	// ChipKind Ascend910/310/310p
	ChipKind string
	// UnhealthyChipIds the card unhealthy chip ids in this node
	UnhealthyChipIds map[int]struct{}
	// ServerType Ascend310p-10-dual cardType-cardCoreNum-duo
	ServerType string
	// TotalChipNum num of total chips, get from capacity
	TotalChipNum int
	// AiCorePerChip num of aicore on each chip
	AiCorePerChip int
	// FreeChipNum num of free chips get from device-info
	FreeChipNum int
	// TotalRes total resource on node
	TotalRes util.VResource
	// ValidVNode node init success
	ValidVNode bool
	// Chip type 910B1/910B2C/910B3/910B4
	ChipType string
}

type VolcanoFrame struct {
	UID        types.UID
	KubeClient kubernetes.Interface

	VJobTemplate map[string]map[string]util.VResource
	ConfigParameters
	// contains filtered or unexported fields
}